Summary: This patch is an implementation of sema and parsing for the OpenMP composite pragma 'distribute simd'.
Differential Revision: http://reviews.llvm.org/D22007
llvm-svn: 274604
- Added new Builtins: enqueue_kernel, get_kernel_work_group_size
and get_kernel_preferred_work_group_size_multiple.
These Builtins use custom check to diagnose parameters of the passed Blocks
i. e. variable number of 'local void*' type params, and check different
overloads specified in Table 6.31 of OpenCL v2.0.
- IR is generated as an internal library call for each OpenCL Builtin,
reusing ObjC Block implementation.
Review: http://reviews.llvm.org/D20249
llvm-svn: 274540
Summary: This patch is an implementation of sema and parsing for the OpenMP composite pragma 'distribute parallel for simd'.
Differential Revision: http://reviews.llvm.org/D21977
llvm-svn: 274530
Currently we only have OpenCL 2.0 Builtins i.e. pipes or address space conversions.
They have to be added only in the version 2.0 compilation mode to make the identifiers
available for use in the other versions.
Review: http://reviews.llvm.org/D20249
llvm-svn: 274509
member is redundantly redeclared outside the class definition in code built in
c++17 mode, ensure we emit a non-discardable definition of the data member for
c++11 and c++14 compilations to use.
llvm-svn: 274416
The CodeView printer expects to be able to generate fully qualified
names from the debug info graph. This means that we need to include the
MSVC-style name in the debug info for anonymous types.
llvm-svn: 274401
With all MaterializeTemporaryExprs coming with a ExprWithCleanups, it's
easy to add correct lifetime.end marks into the right RunCleanupsScope.
Differential Revision: http://reviews.llvm.org/D20499
llvm-svn: 274385
This is important for building libclc. Since r273039 tests are failing
due to now emitting calls to these functions instead of emitting the
DAG node. The libm function names are implemented for OpenCL, and should
call the locally defined versions, so -fno-builtin is used. The IR
Some functions use the __builtins and expect the intrinsics to be
emitted. Without this we end up with nobuiltin calls to intrinsics
or to unsupported library calls.
llvm-svn: 274370
Summary:
This lets LLVM perform IPO over these functions. In particular, it
allows LLVM to emit ld.global.nc for loads to __restrict pointers in
kernels that are never written to.
Reviewers: rsmith
Subscribers: cfe-commits, tra
Differential Revision: http://reviews.llvm.org/D21337
llvm-svn: 274261
When -gmlt is on, we don't emit namespace or class scope information,
and the CodeView emission code in LLVM can't compute the fully qualified
name. If we know LLVM won't be able to get the name right, go ahead and
emit the qualified name in the frontend.
We could change our -gmlt emission strategy to include those scopes when
emitting codeview, but that would increase memory usage and slow down
LTO and add more complexity to debug info emission.
The same problem exists when you debug a -gmlt binary with GDB, so we
should consider removing '&& EmitCodeView' from the condition here at
some point in the future after evaluating the impact on object file
size.
llvm-svn: 274246
Summary:
Summary:
Change Clang calling convention SpirKernel to OpenCLKernel.
Set calling convention OpenCLKernel for amdgcn as well.
Add virtual method .getOpenCLKernelCallingConv() to TargetCodeGenInfo
and use it to set target calling convention for AMDGPU and SPIR.
Update tests.
Reviewers: rsmith, tstellarAMD, Anastasia, yaxunl
Subscribers: kzhuravl, cfe-commits
Differential Revision: http://reviews.llvm.org/D21367
llvm-svn: 274220
Emit the underlying storage offset in addition to the starting bit
position of the field.
This fixes PR28162.
Differential Revision: http://reviews.llvm.org/D21783
llvm-svn: 274201
Reverts r273305 and re-instates r273296.
We needed to fix a bug in Sema::MarkVTableUsed to ensure that operator
delete lookup occurs when the vtable is referenced. We already had a
special case to look up operator delete when dllimport was used, but I
think should really mark virtual destructors referenced any time the
vtable is used.
llvm-svn: 274147
This reverts commit 161ff9db3a3d0d62880d1cb18d58182cd3034912 (r273056).
This is breaking stage2 instrumented builds with "malformed coverage
data" errors.
llvm-svn: 274104
For the new hotness attribute, the API will take the pass rather than
the pass name so we can no longer play the trick of AlwaysPrint being a
special pass name. This adds a getter to help the transition.
There is also a corresponding llvm patch.
llvm-svn: 274101
Replace inheriting constructors implementation with new approach, voted into
C++ last year as a DR against C++11.
Instead of synthesizing a set of derived class constructors for each inherited
base class constructor, we make the constructors of the base class visible to
constructor lookup in the derived class, using the normal rules for
using-declarations.
For constructors, UsingShadowDecl now has a ConstructorUsingShadowDecl derived
class that tracks the requisite additional information. We create shadow
constructors (not found by name lookup) in the derived class to model the
actual initialization, and have a new expression node,
CXXInheritedCtorInitExpr, to model the initialization of a base class from such
a constructor. (This initialization is special because it performs real perfect
forwarding of arguments.)
In cases where argument forwarding is not possible (for inalloca calls,
variadic calls, and calls with callee parameter cleanup), the shadow inheriting
constructor is not emitted and instead we directly emit the initialization code
into the caller of the inherited constructor.
Note that this new model is not perfectly compatible with the old model in some
corner cases. In particular:
* if B inherits a private constructor from A, and C uses that constructor to
construct a B, then we previously required that A befriends B and B
befriends C, but the new rules require A to befriend C directly, and
* if a derived class has its own constructors (and so its implicit default
constructor is suppressed), it may still inherit a default constructor from
a base class
llvm-svn: 274049
[OpenMP] Initial implementation of parse and sema for composite pragma 'distribute parallel for'
This patch is an initial implementation for #distribute parallel for.
The main differences that affect other pragmas are:
The implementation of 'distribute parallel for' requires blocking of the associated loop, where blocks are "distributed" to different teams and iterations within each block are scheduled to parallel threads within each team. To implement blocking, sema creates two additional worksharing directive fields that are used to pass the team assigned block lower and upper bounds through the outlined function resulting from 'parallel'. In this way, scheduling for 'for' to threads can use those bounds.
As a consequence of blocking, the stride of 'distribute' is not 1 but it is equal to the blocking size. This is returned by the runtime and sema prepares a DistIncrExpr variable to hold that value.
As a consequence of blocking, the global upper bound (EnsureUpperBound) expression of the 'for' is not the original loop upper bound (e.g. in for(i = 0 ; i < N; i++) this is 'N') but it is the team-assigned block upper bound. Sema creates a new expression holding the calculation of the actual upper bound for 'for' as UB = min(UB, PrevUB), where UB is the loop upper bound, and PrevUB is the team-assigned block upper bound.
llvm-svn: 273884
http://reviews.llvm.org/D21564
This patch is an initial implementation for #distribute parallel for.
The main differences that affect other pragmas are:
The implementation of 'distribute parallel for' requires blocking of the associated loop, where blocks are "distributed" to different teams and iterations within each block are scheduled to parallel threads within each team. To implement blocking, sema creates two additional worksharing directive fields that are used to pass the team assigned block lower and upper bounds through the outlined function resulting from 'parallel'. In this way, scheduling for 'for' to threads can use those bounds.
As a consequence of blocking, the stride of 'distribute' is not 1 but it is equal to the blocking size. This is returned by the runtime and sema prepares a DistIncrExpr variable to hold that value.
As a consequence of blocking, the global upper bound (EnsureUpperBound) expression of the 'for' is not the original loop upper bound (e.g. in for(i = 0 ; i < N; i++) this is 'N') but it is the team-assigned block upper bound. Sema creates a new expression holding the calculation of the actual upper bound for 'for' as UB = min(UB, PrevUB), where UB is the loop upper bound, and PrevUB is the team-assigned block upper bound.
llvm-svn: 273705
smaller than register as argument in variadic functions on
big endian architectures.
Differential Revision: http://reviews.llvm.org/D21611
llvm-svn: 273665
Summary: InstCombine needs to be performed after simplifycfg and sroa, otherwise it may make bad optimization decisions.
Reviewers: davidxl, wmi, dnovillo
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D21568
llvm-svn: 273606
The PIC and PIE levels are not independent. In fact, if PIE is defined
it is always the same as PIC.
This is clear in the driver where ParsePICArgs returns a PIC level and
a IsPIE boolean. Unfortunately that is currently lost and we pass two
redundant levels down the pipeline.
This patch keeps a bool and a PIC level all the way down to codegen.
llvm-svn: 273566
We would incorrectly emit the directive sections due to the missing overridden
methods. We now emit the expected "/DEFAULTLIB" rather than "-l" options for
requested linkage
llvm-svn: 273558
Push a new region for the try block and propagate execution counts
through it. This ensures that catch statements get a region counter
distinct from the try block's counter.
llvm-svn: 273463
Add support for /Ob1 (and equivalent -finline-hint-functions), which enable
inlining only for functions marked inline, either explicitly (via inline
keyword, for example), or implicitly (function definition in class body,
for example).
This works by enabling inlining pass, and adding noinline attribute to
every function not marked inline.
Patch by Rudy Pons <rudy.pons@ilod.org>!
Differential Revision: http://reviews.llvm.org/D20647
llvm-svn: 273440
It currently only takes 2048 gotos to overflow the FixupDepth bitfield,
causing silent miscompilation. Apparently some parser generators run into
this (see PR).
I don't know that that data structure is terribly size sensitive anyway,
and since there's no room to widen the bitfield, let's just use a separate
word in EHCatchScope for it.
Differential Revision: http://reviews.llvm.org/D21566
llvm-svn: 273434
This patch uses function metadata to represent reqd_work_group_size, work_group_size_hint and vector_type_hint kernel attributes and kernel argument info.
Differential Revision: http://reviews.llvm.org/D20979
llvm-svn: 273425
Summary:
If the RenderScript LangOpt is set, either via '-x renderscript' or the '.rs'
file extension, set the DWARF language tag to be that of RenderScript.
Reviewers: rsmith
Subscribers: cfe-commits, srhines
Differential Revision: http://reviews.llvm.org/D21451
llvm-svn: 273321
Summary: We need to call PruneEH pass before AutoFDO pass so that some EH-related calls can get inlined in Sample Profile pass.
Reviewers: davidxl, dnovillo
Subscribers: junbuml, llvm-commits
Differential Revision: http://reviews.llvm.org/D21197
llvm-svn: 273298
MSVC doesn't provide them. PR28223
I left behind the machinery in case we want to resurrect available_externally
vftable emission to support devirtualization.
Reviewers: majnemer
Differential Revision: http://reviews.llvm.org/D21544
llvm-svn: 273296
The export side is responsible for running any initializers, they are
run when the module is first loaded. Attempting to run an initializer
for the import side is not possible.
This fixes PR28216.
llvm-svn: 273237
Given something like:
void *v = (void *)100;
We need to synthesize a ptrtoint operation from 100. During constant
emission, we choose i64 as the type for our constant because it
guaranteed not to drop any bits from our CharUnits representation of the
value. However, this is suboptimal for 32-bit targets: LLVM passes like
GlobalOpt will get confused by these sorts of casts resulting in
pessimization.
Instead, make sure the ptrtoint operand has a pointer-sized integer
type.
llvm-svn: 273020
Reapplying patch in r272777 which was reverted
because the llvm patch which added support
for generating the mcrr/mcrr2 instructions
from the intrinsic was causing an assertion
failure. This has now been fixed in llvm.
llvm-svn: 272983
This patch fixes a bug where we'd segfault (in some cases) if we saw a
variadic function with one or more pass_object_size arguments.
Differential Revision: http://reviews.llvm.org/D17462
llvm-svn: 272971
Summary:
This patch fixes an issue detected when firstprivate variables are passed to an OpenMP outlined function vararg list. Currently they are not compatible with what the runtime library expects causing malfunction in some targets.
This patch fixes the issue by moving the casting logic already in place for offloading to the common code that creates the outline function and arguments and updates the regression tests accordingly.
Reviewers: hfinkel, arpith-jacob, carlo.bertolli, kkwli0, ABataev
Subscribers: cfe-commits, caomhin
Differential Revision: http://reviews.llvm.org/D21150
llvm-svn: 272900
Parameters and non-static members of aggregates are still excluded,
and probably should remain that way.
Differential Revision: http://reviews.llvm.org/D19754
llvm-svn: 272859
As noted in the code comment, a potential follow-on would be to remove
the builtins themselves. Other than ord/unord, this already works as
expected. Eg:
typedef float v4sf __attribute__((__vector_size__(16)));
v4sf fcmpgt(v4sf a, v4sf b) { return a > b; }
Differential Revision: http://reviews.llvm.org/D21268
llvm-svn: 272840
Patch adds intrinsics for mrrc/mrrc2. The
intrinsics for mrrc/mrrc2 return a single
uint64_t to represent two 32 bit values.
The mcrr/mcrr2 intrinsic was changed to
accept a single uint64_t instead of two
32 bit values as the input for consistency.
Differential Revision: http://reviews.llvm.org/D21179
llvm-svn: 272777
classes.
MSVC actively uses unqualified lookup in dependent bases, lookup at the
instantiation point (non-dependent names may be resolved on things
declared later) etc. and all this stuff is the main cause of
incompatibility between clang and MSVC.
Clang tries to emulate MSVC behavior but it may fail in many cases.
clang could store lexed tokens for member functions definitions within
ClassTemplateDecl for later parsing during template instantiation.
It will allow resolving many possible issues with lookup in dependent
base classes and removing many already existing MSVC-specific
hacks/workarounds from the clang code.
llvm-svn: 272774
Summary:
This is similar to other loop pragmas like 'vectorize'. Currently it
only has state values: distribute(enable) and distribute(disable). When
one of these is specified the corresponding loop metadata is generated:
!{!"llvm.loop.distribute.enable", i1 true/false}
As a result, loop distribution will be attempted on the loop even if
Loop Distribution in not enabled globally. Analogously, with 'disable'
distribution can be turned off for an individual loop even when the pass
is otherwise enabled.
There are some slight differences compared to the existing loop pragmas.
1. There is no 'assume_safety' variant which makes its handling slightly
different from 'vectorize'/'interleave'.
2. Unlike the existing loop pragmas, it does not have a corresponding
numeric pragma like 'vectorize' -> 'vectorize_width'. So for the
consistency checks in CheckForIncompatibleAttributes we don't need to
check it against other pragmas. We just need to check for duplicates of
the same pragma.
Reviewers: rsmith, dexonsmith, aaron.ballman
Subscribers: bob.wilson, cfe-commits, hfinkel
Differential Revision: http://reviews.llvm.org/D19403
llvm-svn: 272656
pointer-to-pointer representing the parameter. An aggregate rvalue representing
a pointer does not make sense.
We got away with this weirdness because CGCall happens to blindly load an
RValue in aggregate form in this case, without checking whether an RValue for
the type should be in scalar or aggregate form.
llvm-svn: 272609
We can now use __builtin_nontemporal_store instead of target specific builtins for naturally aligned nontemporal stores which avoids the need for handling in CGBuiltin.cpp
The scalar integer nontemporal (unaligned) store builtins will have to wait as __builtin_nontemporal_store currently assumes natural alignment and doesn't accept the 'packed struct' trick that we use for normal unaligned load/stores.
The nontemporal loads require further backend support before we can safely convert them to __builtin_nontemporal_load
Differential Revision: http://reviews.llvm.org/D21272
llvm-svn: 272540
These ExprWithCleanups are added for holding a RunCleanupsScope not
for destructor calls; rather, they are for lifetime marks. This requires
ExprWithCleanups to keep a bit to indicate whether it have cleanups with
side effects (e.g. dtor calls).
Differential Revision: http://reviews.llvm.org/D20498
llvm-svn: 272296
Summary:
This should have been a very simple change, but it was greatly
complicated by the construction of new Decls during IR generation.
In particular, we reconstruct the AST function type in order to get the
implicit 'this' parameter into C++ method types.
We also have to worry about FunctionDecls whose types are not
FunctionTypes because CGBlocks.cpp constructs some dummy FunctionDecls
with 'void' type.
Depends on D21114
Reviewers: aprantl, dblaikie
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D21141
llvm-svn: 272198
__builtin_astype does not generate correct LLVM IR for vec3 types. This patch inserts bitcasts to/from vec4 when necessary in addition to generating vector shuffle. Sema and codegen tests are added.
Differential Revision: http://reviews.llvm.org/D20133
llvm-svn: 272153
We have an assertion failure if, for example, the definition of an unused
inline function starts in one macro and ends in another. This patch fixes
the issue by finding the common ancestor of the start and end locations
of that function's body and changing the locations accordingly.
Thanks to NAKAMURA Takumi for helping with fixing the test failure on Windows.
Differential Revision: http://reviews.llvm.org/D20997
llvm-svn: 271995
We have an assertion failure if, for example, the definition of an unused
inline function starts in one macro and ends in another. This patch fixes
the issue by finding the common ancestor of the start and end locations
of that function's body and changing the locations accordingly.
Differential Revision: http://reviews.llvm.org/D20997
llvm-svn: 271969
The assertion added earlier was overly strict. We need to strip the pointer
casts (as when constructing the GV). Correct the types (Function or Variable).
llvm-svn: 271750
The `isa' member was previously not given the correct DLL Storage. Ensure that
we give the `isa' constant `__CFConstantStringClassReference' the correct DLL
storage. Default to dllimport unless an explicit specification gives it a
dllexport storage.
llvm-svn: 271361
Summary:
This is particularly important because a some convergent CUDA intrinsics
(e.g. __shfl_down) are implemented in terms of inline asm.
Reviewers: tra
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D20836
llvm-svn: 271336
I added this call in r271308. It's redundant because it's dominated by a
call to extendRegion().
Thanks to Justin Bogner for pointing this out!
llvm-svn: 271331
Craig Topper